Online Program Home
My Program

Abstract Details

Activity Number: 582 - Nonparametric Methods for Statistical Inference
Type: Contributed
Date/Time: Wednesday, July 31, 2019 : 2:00 PM to 3:50 PM
Sponsor: Section on Nonparametric Statistics
Abstract #307332 Presentation
Title: A Generalized Additive Cox Model with L1-Penalty for Heart Failure Time-To-Event Outcomes and Comparison to Other Machine Learning Approaches
Author(s): Matthias Kormaksson*
Keywords: GAM; LASSO; Cox Regression; R; Heart Failure

The original objective of this work was to assess performance of three survival analysis methods for predicting time to event in heart failure studies. The three methods were Cox regularized regression (Cox-LASSO), generalized additive Cox regression (Cox-GAM), and random survival forest, which provides a popular alternative from the Machine Learning community. The first two methods have the attractive properties of allowing for variable selection (Cox-LASSO) and modeling of non-linear features (Cox-GAM). However, currently there is no method that can incorporate both of these desirable properties. Therefore, we developed a novel hybrid method involving generalized additive Cox regression with additional L1-penalty (Cox-GAMLASSO), which has been submitted to CRAN as an R-package. All models were trained and tested on data from two large studies in the Novartis Heart Failure database that includes both clinical and biomarker data. Model comparisons were made based on both discrimination and calibration performance. We conclude with lessons learned about the advantages and disadvantages of the different Statistical & Machine Learning methods.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program